上海交通大学学报(自然版) ›› 2013, Vol. 47 ›› Issue (01): 81-85.

• 自动化技术、计算机技术 • 上一篇    下一篇

采用数据驱动机制的多核处理器

毕卓a,徐云川b,王镇b   

  1.  (上海大学a.机电工程与自动化学院自动化系; b.微电子研发中心, 上海 200072)
  • 收稿日期:2012-05-20 出版日期:2013-01-30 发布日期:2013-01-30

Data Driven Multi-core Processor

 BI  Zhuo-a, XU  Yun-Chuan-b, WANG  Zhen-b   

  1. (a.School of Mechatronic Engineering and Automation; b.Microelectronics R&D Center, Shanghai University, Shanghai 200072, China)
  • Received:2012-05-20 Online:2013-01-30 Published:2013-01-30

摘要: 摘要: 
针对当前工艺条件下多核处理器存在程序并行性不足的问题,设计了一种采用数据驱动机制、支持函数语言风格编程的多核处理器,包括通用处理器核、数据驱动模块和片内路由器.其中:通用处理器核用于执行常规程序;数据驱动模块用于检测数据的完备性;片内路由器则可提供处理器核之间及簇之间的通信.实验结果表明,所设计的多核处理器能够支持C语言“函数式语言”风格的编程模板.每个C代码段执行纯函数的操作,消除了函数间的共享变量,使得并行编程的复杂度有所降低.同时,所采用的数据驱动机制没有执行顺序的严格限制,充分挖掘了算法潜在的并行性.经测试,数据驱动多核处理器的加速比随着计算资源的增加而增大,从而验证了数据流计算机的加速倍数随处理器数目增加而线性增长的结论.
关键词: 
数据驱动; 并行编程模型; 数据流机; 多核处理器
中图分类号:  TP 338
文献标志码:  A    

Abstract:  For the problem on insufficient parallelism in program of multi-core processors encountered in current technology, a multi-core processor prototype was presented. It supports data driven mechanism and programming style of functional language. The prototype consists of general processor cores, data driven module and router on chip. General processor cores are used to execute normal operation. Data driven module detects the data completeness. Router on chip provides communication among processor cores or clusters. The experimental results show that multi-core processor prototype supports the functional language programming style of C language. The execution of every C code segment is the operation of pure function, which eliminates the sharing variables among functions and partly reduces the complexity of parallel program. Moreover, there are no strict sequential restrictions over the data driven mechanism, which fully excavates the potential parallelism of algorithms. According to the test, the multi-core processors based on data driven mechanism can obtain linear acceleration ratio with the increase of computation source. It preliminarily proves that the acceleration ratio of data flow machine maintains linearly increasing with the increase of processors.

Key words: data driven, parallel programming model, data flow machine, multi-core processors